feat: project sessions, notes, and per-repo branch timelines#244
Draft
feat: project sessions, notes, and per-repo branch timelines#244
Conversation
- Add ProjectNote model with project-scoped notes for cross-cutting context - Add project_notes table to SQLite schema with cascade delete - Add reason column to project_repos for tracking why repos are attached - Add start_project_session command for project-level sessions - Add create_project_note, list_project_notes, delete_project_note commands - Build project session context with repo listing, existing notes, and chronological branch timeline summaries grouped by repository - Inject project notes into branch-level timeline context - Add project note extraction to session_runner post-completion hooks - Add ProjectNote and ProjectSessionResponse types to frontend - Add project session UI with prompt input, notes display, and deletion - Bump schema version (wipe-and-recreate on mismatch)
… section - Remove createNote parameter from startProjectSession, always pass true so every project session generates a note - Remove session running state from prompt input — the textarea is always available, allowing multiple concurrent project sessions - Track active sessions in a Set<string> instead of a single ID - Reload notes immediately after starting a session so the stub appears as a 'Generating note…' card with a spinner in the notes section - Support multiple concurrent generating notes via filter instead of find - Remove unused Loader2 import, .running CSS state, and spin animation
… add_project_repo tools Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
These files were written locally but never committed before the rebase moved the directory from staged/ to apps/mark/, leaving the pub mod declarations in lib.rs dangling. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Fix project sessions appearing completed immediately by saving the initial note stub with empty title/content (matching branch note behavior) so the generating state is correctly detected - MCP add_project_repo now blocks until worktree is set up, detect_actions has run, and all prerun setup actions have completed - Emit project-repo-added Tauri event from MCP handler so ProjectHome reactively reloads when a repo is added via MCP without manual refresh - Remove debug log statements from project_mcp.rs, session_commands.rs, session_runner.rs, ProjectSection.svelte, and acp-client/driver.rs Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- start_repo_session: add `return_info` param so parent session can specify what the child should return; child prompt is suffixed with the instruction; last assistant message is returned as `output` in the result JSON - add_project_repo: emit `project-repo-added` Tauri event at each stage (repo created, worktree ready, prerun actions done) so the UI updates progressively without requiring a manual reload Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ards When a project session adds a repo via MCP with a reason, save it to the DB and display it as the first item in the branch card timeline — styled with an info (blue) color theme and an Info icon. Hovering reveals an X button on the right to dismiss it, which clears the reason from the DB. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Before creating the repo, run the monorepo check. If the score is >= 20 and no subpath was provided, return an error telling the agent it must re-call the tool with a subpath pointing to the specific service or package root (e.g. "packages/api"). Also expose `subpath` as an explicit MCP parameter so agents can supply it directly. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…+subpath Add subpath parameter to StartRepoSessionParams and update the repo lookup to match on both github_repo and subpath exactly, so multiple copies of the same repo with different subpaths are disambiguated correctly. Calls now fail immediately with a clear error listing available repo+subpath combos if no match is found. Also sets the working directory to include the subpath. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…d unread tracking Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the implicit return-only behaviour with an explicit `expected_outcome` field (`return_output_only`, `note_in_repo`, `commit`). When `note_in_repo` or `commit` is requested the agent receives the same action-prefix prompt used by branch sessions, a Note or Commit stub is created in the DB (linked via session_id so the branch card timeline picks it up automatically), and commit sessions capture the pre-HEAD SHA so the post-completion hook can resolve the new SHA. Expand the log line to include all params including the full instructions. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…h cards live When start_repo_session creates a note/commit artifact, it now emits a session-status-changed event with status "running" and branch_id, project_id, session_type fields. BranchCard listens for running events matching its branch and calls loadTimeline() immediately, so the pending stub appears without a reload. App.svelte registers the session in sessionRegistry and projectStateStore so the project spinner activates and cleanup on completion works correctly. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
start_repo_session now builds the same <branch-history> and project information context that user-triggered branch sessions receive. Looks up the worktree path for local branches (falling back to the clone path) and calls the now pub(crate) build_branch_context / build_remote_branch_context helpers from session_commands. The prompt structure matches build_full_prompt: <action> block with action instructions + project info, <branch-history> block, then user instructions. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace the blanket `git fetch origin` in `ensure_local_clone` with a new `fetch_for_worktree` helper that fetches only `base_branch` and `branch_name`. The full fetch updated every remote-tracking ref via compare-and-swap, which caused spurious "incorrect old value provided" failures (and a needless HTTPS retry) when another worktree or process had already updated some refs in the shared clone. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…tart_repo_session call Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
base_branch is stored in the DB with an "origin/" prefix but git fetch expects the bare ref name. Without stripping, fetch ran: git fetch origin origin/main which failed with "couldn't find remote ref origin/main". Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…o on completion - Clarify note_in_repo and commit expected_outcome descriptions with usage guidance - Explain that existing notes are accessible by name in the instructions field - Add guidance on reason field for add_project_repo (describe repo, not todos) - Return created note info in the same format as session-start notes when a note_in_repo session completes Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…ponse The note is now returned separately in the `note` field, so the raw agent output no longer needs to include the content after the `---` separator. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…essions Ensures project sessions always have a real working directory by creating the project-scoped worktree root (~/.staged/workspaces/local/projects/<id>/) at project creation time, and falling back to it instead of /tmp when the project has no repos attached yet. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…f-race errors fetch_for_worktree already fetches the base branch with SSH→HTTPS fallback before create_worktree_at_path is called, so the second best-effort fetch inside create_worktree_at_path was redundant and could cause duplicate ref-update races. Also swallow "incorrect old value provided" errors in fetch_for_worktree: this is a CAS race where a concurrent fetch already updated the tracking ref, leaving it at least as up-to-date as needed, so the error is non-fatal. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
start_repo_sessionandadd_project_repotools, enabling AI agents to coordinate work across multiple repos in a projectcreate_worktree_at_path(superseded byfetch_for_worktree) and swallows git CAS race errors (incorrect old value provided) that were incorrectly treated as fatalTest plan
add_project_repoMCP tool correctly rejects monorepo adds without a subpath🤖 Generated with Claude Code